Automatic gain control filter in a video analysis system
Techniques are disclosed for analyzing a scene depicted in an input stream of video frames captured by a video camera. Bounding boxes are determined for a set foreground patches identified in a video frame. For each bounding box, the techniques include determining textures for first areas, each including a foreground pixel and surrounding pixels, and determining textures for second areas including pixels of the background model image corresponding to the pixels of the foreground areas. Further, for each foreground pixel in the bounding box area, a correlation score is determined based on the texture of the corresponding first area and second area. Pixels whose correlation scores exceed a threshold are removed from the foreground patch. The size of the bounding box may also be reduced to fit the modified foreground patch.
1. A computer-implemented method comprising:
computing at least one bounding box for one or more foreground objects in a video frame;
for each bounding box:
determining a texture of a first area including pixels for the one or more foreground objects and pixels surrounding the pixels for the one or more foreground objects and a texture of a second area including pixels of a background model image for the video frame corresponding to pixels of the first area,
determining a correlation score based on the texture of the first area and the texture of the second area, and
resizing the bounding box based on the correlation score and a threshold value, the threshold value being proportional to a size of the bounding box; and
outputting the video frame with the resized at least one bounding box for behavior analysis of the one or more foreground objects.
2. The method of claim 1 , wherein determining texture of the first area of an associated bounding box includes determining gradient values for the pixels for the one or more foreground objects and the pixels surrounding the pixels for the one or more foreground objects, and wherein determining texture of the second area includes determining gradient values for the pixels of the background model image corresponding to the pixels of the first area.
3. The method of claim 2 , wherein the gradient values are determined using an edge detection technique which is resilient to noise and invariant to illumination and/or color/hue changes.
4. The method of claim 2 , wherein the gradient values are determined by applying one of a Sobel Operator and a Haar wavelet transform.
5. The method of claim 1 , wherein the number of surrounding pixels used in the first area is proportional to the size of the associated bounding box.
6. The method of claim 1 , further comprising, removing the pixels for the one or more foreground objects from the first area of the associated bounding box when the correlation score exceeds the threshold value.
7. The method of claim 1 , further comprising, prior to determining the textures, converting the pixels in the first area and the second area of the associated bounding box to grayscale.
8. A non-transitory computer-readable storage medium storing computer-executable instructions, which when executed by a computer system, perform operations, the computer-executable instructions comprising instructions to:
compute at least one bounding box for foreground objects in a video frame;
for the at least one bounding box:
determine a texture of a first area including pixels for the foreground objects and pixels surrounding the pixels for the foreground objects and a texture of a second area including pixels of a background model image for the video frame corresponding to pixels of the first area, and
determine a correlation score based on the texture of the first area and the texture of the second area; and
resize the at least one bounding box based on the correlation score and a threshold value, the threshold value being proportional to a size of the at least one bounding box; and
output the video frame with the resized at least one bounding box to analyze behavior of the foreground objects.
9. The non-transitory computer-readable storage medium of claim 8 , wherein the computer-executable instructions to determine texture of the first area include instructions to determine gradient values for the pixels for the foreground objects and the pixels surrounding the pixels for the foreground objects, and wherein the computer-executable instructions to determine texture of the second area includes instructions to determine gradient values for the pixels of the background model image corresponding to the first area.
10. The non-transitory computer-readable storage medium of claim 9 , wherein the instructions to determine gradient values include instructions for an edge detection technique that is resilient to noise and invariant to illumination and/or color/hue changes.
11. The non-transitory computer-readable storage medium of claim 9 , wherein the instructions to determine gradient values include instructions for application of at least one of a Sobel Operator and a Haar wavelet transform.
12. The non-transitory computer-readable storage medium of claim 8 , wherein the number of surrounding pixels used in the first area is proportional to the size of the at least one bounding box.
13. The non-transitory computer-readable storage medium of claim 8 , wherein the computer-executable instructions further comprises instructions to remove the pixels for the foreground objects from the first area when the correlation score exceeds the threshold value.
14. The non-transitory computer-readable storage medium of claim 8 , the computer-executable instructions further comprising instructions to, prior to determining the textures, convert the pixels in the first area and the second area to grayscale.
15. A system, comprising:
at least one processor; and
a memory in communication with the at least one processor, wherein the memory includes one or more application programs configured to perform operations, the one or more application programs comprising instructions for the at least one processor to:
compute at least one bounding box for one or more foreground objects in a video frame;
for each bounding box:
determine a texture of a first area including pixels for the one or more foreground objects and pixels surrounding the pixels for the one or more foreground objects and a texture of a second area including pixels of a background model image for the video frame corresponding to pixels of the first area,
determine a correlation score based on the texture of the first area and the texture of the second area, and
resize the bounding box based on the correlation score and a threshold value, the threshold value being proportional to a size of the at least one bounding box; and
output the video frame with the resized at least one bounding box for behavior analysis of the one or more foreground objects.